第七届CSIG图像图形技术挑战赛—复杂工业场景异常检测算法挑战赛

Who this competition fits

Learners comfortable reading Chinese problem statements who want industry or algorithm challenges on Tianchi.

What this competition gives you

There is a baseline you can run as-is — starting from it is usually far faster than building from scratch.

• DataReal-IAD Variety deep-cleaned subset: 100 categories and over 31,000 files.
• DataDINOv2 pretrained weights repository.
• BaselineINP-Former open-source code repository.
• DataOpenCLIP cross-modal toolkit repository.
• Reference paperReal-IAD CVPR 2024 prior-work reference.
• Reference paperReal-IAD D³ CVPR 2025 prior-work reference.
• Reference paperReal-IAD Variety PR 2026 prior-work reference.
• Reference paperWinCLIP prior-work reference for zero-shot defect inference.
• Reference paperAnomalyGPT prior-work reference for zero-shot defect inference.
• Reference paperAdaCLIP prior-work reference for zero-shot defect inference.
• Support channelCompetition support email.
• Support channelCompetition discussion group.

From registration to a first submission

1. 01

   Python basics

2. 02

   Transformer or LLM workflow basics

3. 03

   prompting or evaluation workflow

4. 04

   competition submission workflow

Before you commit: Task framing, data access, and evaluation details live mainly on the official page and may require careful reading.